Moonshot AI is in early talks with Microsoft, Amazon, and Google to host its Kimi K3 model on their clouds under revenue-sharing deals that could give Moonshot up to 30% of the revenue those services generate. K3, a 2.8 trillion-parameter open-weight model, is too compute-intensive for most organizations to self-host, so cloud distribution matters despite its open license. Key unresolved issues include the revenue split, data access provisions, and how token usage would be audited, complicated by data-privacy concerns given Moonshot is a Chinese company. Politics loom large: US Treasury Secretary Scott Bessent has floated blacklisting Moonshot over allegations of model distillation and illegal chip acquisition, which Moonshot denies. The timing matters commercially for Moonshot, which is preparing a Hong Kong listing at a $30bn valuation after a sevenfold rise in six months.
Questions this post answers
How many parameters does Moonshot's Kimi K3 model have?
Kimi K3 has 2.8 trillion parameters, making it the largest open-weight model released to date. Its size makes it open in license but impractical for most organizations to self-host, since the compute needed to serve it exceeds what all but a handful of companies can assemble, which is why cloud distribution deals matter. Track how massive open-weight models like Kimi K3 reshape cloud hosting economics on daily.dev.
What revenue share is Moonshot AI reportedly asking US cloud providers for hosting Kimi K3?
Moonshot AI is reportedly seeking up to 30% of the revenue Microsoft, Amazon, and Google would generate from hosting Kimi K3, a figure said to match what it has asked other large customers. This inverts the usual arrangement where the cloud platform takes the cut and the model developer keeps the remainder. Follow unusual AI licensing and revenue-share deals shaping cloud strategy on daily.dev.
Why might US cloud providers hesitate to host a Chinese AI model like Kimi K3?
Hosting Kimi K3 on Azure, AWS, or Google Cloud raises data access concerns, since a Chinese company hosting the model could see prompts and outputs, worrying enterprise customers with sensitive data. Political pressure adds risk too: US Treasury Secretary Scott Bessent has suggested adding Moonshot to trade blacklists over distillation and chip-acquisition allegations Moonshot denies. Keep up with the data-access and policy risks around foreign AI models via daily.dev.
Share this post